More than 100 years after it was first described, Alzheimer's disease (AD) is still diagnosed strictly on clinical criteria that are not sensitive to the early stages of disease and do not adequately distinguish AD from non-AD dementias. When comparing a clinical diagnosis to an autopsy-confirmed diagnosis, even the most seasoned clinicians are wrong 15-20% of the time. This diagnostic inaccuracy is assumed to be considerably worse in non-specialty settings where the initial diagnosis of AD is most often made. With advances in experimental therapeutics and the concomitant reminder that potent treatments may have serious side effects, the need for an accurate, non-invasive AD biomarker is more pressing than ever. Such a biomarker would be useful on several fronts. In the clinical setting it would allow for more diagnostic certainty in trying to distinguish AD from other dementias. An accurate biomarker with sufficient sensitivity should also help predict which patients with mild cognitive impairment (MCI) will go on to develop AD and, just as importantly, which will not. Lastly, an AD biomarker that detects disease in the earliest stages and tracks with clinical status would accelerate drug development by facilitating dose-response studies and enabling more rapid and objective assessment of efficacy. Despite considerable efforts, the field has yet to develop a biomarker that can meet these pressing needs. The current application will examine a relatively novel form of functional MRI (fMRI) as a candidate imaging biomarker in AD. Resting-state fMRI provides a measure of functional connectivity within specific brain networks and has shown promise in preliminary studies as an AD biomarker. The limitations of this approach currently are that it has not yet proven to be reliably interpretable at the single-subject level, its predictive value in MCI remains uncertain, and it has not been examined longitudinally. The current application, drawing on the strengths of a multi-site, longitudinal study, will attempt to address these limitations. The study will involve the acquisition of resting-state fMRI data from Stanford University and the University of California, San Francisco in four large cohorts of subjects: healthy aging, MCI, AD, and non-AD dementia. Subjects will be scanned at baseline and followed longitudinally. A subset of subjects will be scanned again at a 1-year interval. The aims of the study will be a) to enhance the sensitivity and specificity of resting-state functional connectivity measures in distinguishing AD from both healthy aging and non-AD dementia, b) to assess the utility of resting-state fMRI in predicting which patients with MCI subsequently convert to AD over the five-year course of the study and c) to assess the utility of resting-state fMRI in tracking disease progression over time.